DocsWriting Studio › Which notes the AI sees, and which it can write

Which notes the AI sees, and which it can write

Last updated September 9, 2026 · 7 min read

In one line: which notes reach the AI without you asking, which it has to go looking for, and what it can and cannot write into your notebook.

Two channels, kept apart

Notes reach the AI two different ways. Confusing them is the usual reason a note seems to have been ignored.

Sent without asking

  • Notes on the file you have open
  • Your book-level notes
  • Attached to every message you send

It has to go looking

  • Notes on other chapters
  • Notes you marked done
  • The full body of a book note

What is sent without asking

When you send a message from the AI Chat panel, Slima attaches your notes to it as references:

  • Chapter Notes — the notes on the file you have open right now that are not marked done, title and full body.
  • Book Notes — your book-level notes that are not marked done, each as a title plus the opening of its body.

Each arrives labelled, so you can see them attached in the conversation. Messages the coach starts on its own carry the same two bundles.

Two useful consequences:

  1. Marking a note done takes it out of the conversation. Done notes are not attached. When you have settled a thread, mark it done and the AI stops circling back to it.
  2. Book notes are sent as an excerpt. So put the operative sentence of a book note first, not in the third paragraph. Chapter notes go in whole, which means a long chapter note really does get read.

The flip side: with a file closed, its notes are not in the conversation. To talk about the setup in chapter 8, open chapter 8 — or just ask the AI to list your notes.

What it has to go looking for

Everything else needs a tool. The AI has two read-only ones:

  • List notes — its only way to discover that notes exist. It gets each note's id, title, category, scope and a short preview, and can ask for one category at a time (note / to-do / setup / payoff).
  • Read note — opens one note in full, by id.

There is a threshold on those two: the book has to hold at least one note before they appear in the AI's toolbox. A brand-new book has nothing to list, so they are not offered.

Can it create notes on its own

It can, and by default it won't.

The create-note tool is always in its hands — even in an empty book, or the first note could never be made. But the standing instruction is coach first: while you are exploring, discussing or stuck, it should ask, offer options and push back, and it should not helpfully produce a summary or a notebook of its own. If you want a note left behind, say so.

When you do ask, it can set:

Field Notes
Title Required
Body Optional, Markdown
Category Note / To-do / Setup / Payoff
Scope The book, or one file (it has to supply a path)
Colour The four in the panel
Tags Optional

A few behaviours worth knowing:

  • A setup it creates always starts as not yet paid off.
  • To attach a note to a chapter it must supply a path that resolves. If the path is wrong the call fails, rather than leaving an orphan note pointed at a file that does not exist.
  • If it names a colour the panel does not offer, the note comes out gray instead of failing.
  • A note it creates shows up in your panel immediately.

What it can and cannot change

The edit-note tool can change an existing note's title, body, category, colour and tags — and only the fields it sends; the rest are left alone.

Three things it cannot do:

  • It cannot move a note's scope. Whether a note belongs to the book or to a file is decided when the note is created; the AI has no field for it.
  • It cannot change the category of a paired setup or payoff. Ask it to, and it refuses and tells you to unlink the pair in the panel first — unlinking is your call, not its.
  • It cannot delete a note. There is no path by which the AI removes a note of yours.

Editing a body replaces it; it does not append. Ask it to "tidy up this note" and the wording you had is gone. If you want to keep the old version, copy it first, or ask for the new one in chat and paste it yourself.

Notes are the one AI write that skips review

When the AI changes your manuscript, the change arrives as a proposal — a dashed card you accept or skip (see Change review for AI edits). Notes do not work that way. Creating and editing a note lands in the panel straight away: no preview, no accept button.

So after you ask it to touch a note, read what it wrote. Slima does take an automatic version snapshot of the book before the tool runs — but notes are not in that snapshot. They are not under version control, so a note body it overwrote has no history to restore from.

Requests that actually get you something

  • "Read chapters 1 to 8 and tell me which setups I never paid off, then record what you find as setup and payoff notes." It will read, and it will create. The pairing click is yours — it has no pairing tool.
  • "Turn what we just decided into a to-do on chapter 3."
  • "List my book notes and pick out the ones that contradict each other." That one uses both read-only tools.
  • "In the setup note on chapter 2, make the 'do not explain yet' part more specific."

Your notebook is not its scratchpad

What the AI remembers about your book is a separate thing and does not live in your notes. Notes appear in your interface as your own thinking, so working state has no business there. Anything it needs to carry between sessions has somewhere else to go.

An AI tool connected from outside

Connect an outside AI writing tool to a book over MCP and the shape is the same: it can list your notes, read one and leave one, and it cannot delete them. Setup is in What is Slima MCP.

Related

Try it in Slima

Open the app and do this with your own book. Free to start, no credit card.

Open Slima
Was this helpful?